* [ltt-dev] [RFC UST and LTTNG] Daemon model proposal v0.4
2011-02-01 19:26 [ltt-dev] [RFC UST and LTTNG] Daemon model proposal v0.4 David Goulet
2011-02-01 19:26 ` David Goulet
2011-02-01 19:26 ` David Goulet
@ 2011-02-02 8:35 ` Nils Carlson
2011-02-04 16:41 ` David Goulet
2 siblings, 1 reply; 5+ messages in thread
From: Nils Carlson @ 2011-02-02 8:35 UTC (permalink / raw)
Hi David,
I will try to comment as well as I can below...
David Goulet wrote:
> Hi everyone,
>
> Introduce a complete new proposal for LTTng and UST tracing
> infrastructure and control.
>
> This is the fourth revision (v0.4) of that proposal. This version also
> introduce the concept of a global and per-user registry. See the
> changelog at the beginning of the file.
>
> Hope I did not forget anything and sorry if I did... the last replies
> for the v0.3 were quite enormous.
>
> Comments from Mathieu Desnoyers and Michel Dagenais were added. Nothing
> final, we are still in the development process.
>
> Again, questions and comments are more than welcome!
>
> Cheers!
> David
>
> RFC - New processes model for UST and LTTng
>
> Author: David Goulet <david.goulet at polymtl.ca>
>
> Contributors:
> * Mathieu Desnoyers <mathieu.desnoyers at efficios.com>
> * Yannick Brosseau <yannick.brosseau at polymtl.ca>
> * Nils Carlson <nils.carlson at ericsson.com>
> * Michel Dagenais <michel.dagenais at polymtl.ca>
> * Stefan Hajnoczi <stefanha at gmail.com>
>
> Version:
> - v0.1: 17/01/2011
> * Initial proposal
>
> - v0.2: 19/01/2011
> After multiple reply from all the contributors above, here is
> the list
> of what has changed:
> * Change/Add Terminology elements from the initial model
> * New figures for four new scenarios
> * Add inprocess library section
> * LTTng kernel tracer support proposition
> * More details for the Model and Components
> * Improve the basic model. Quite different from the last one
>
> - v0.3: 28/01/2011
> In response from Michel Dagenais and Nils Carlson comments:
> * Add scaling to reasons of this re-engineering
> * Purpose of the session ID
> * Explain why ltt-sessiond creates the tracing buffers
> * ust-consumerd interaction schema
> * Clarify inprocess library behavior
>
> - v0.4: 01/02/2011
> After Mathieu Desnoyers and Michel Dagenais comments:
> * Add section Introduction
> * Define the global and per-user ltt-sessiond
> * Add details for ltt-sessiond in the inprocess lib section
> * Session ID are now UUID
> * Add buffer snapshot schema for ust-consumerd
> * ltt-sessiond validate inprocess lib version
> * ltt-sessiond socket validation by the inprocess lib.
> * Add drtrace definition
> * Add consumer to the Model section
>
> Terminology
> -----------------
>
> ltt-sessiond - Main daemon for trace session registry for UST and LTTng
>
> ust-consumerd - Daemon that consume UST buffers for a specific
> application
>
> ltt-consumerd - Daemon that consume LTTng buffers
>
> tracing session - A trace linked to a set of specific tracepoints and
> to a set
> of tracing buffers
>
> tracing buffers - Buffers containing tracing data
>
> tracing data - Data created by tracing an application
>
> inprocess library - UST library linked with the application
>
> shared memory - system V shared memory
>
> application common named pipe - Global named pipe that triggers
> application
> registration, on pipe event, to
> ltt-sessiond
>
> drtrace - New command line tool for LTTng and UST tracing control
>
> Introduction
> -----------------
>
> This RFC propose a brand new UST and LTTng daemon model. This
> re-engineering
> was mostly driven by the need of:
>
> * Better security in terms of access rights on tracing data
> * Manage tracing session
> * Scaling in terms of thread/processes needed to perform tracing
> * LTTng and UST integration in terms of merging traces and session
> control
> * Networking such as streaming and remote control over different
> traces
>
> The new model follows the basic principles of having a session registry
> (ltt-sessiond) and consumers for each tracing session (ust-consumerd and
> ltt-consumerd).
>
> With this proposal, LTTng and UST will share the same tracing session, be
> managed by the same tool and bring a complete integration between
> these two
> powerful tools.
>
> NOTE: This proposal does NOT makes UST dependent on LTTng and vice versa.
>
> Model
> -----------------
>
> A global and/or per-user registry keeps track of all tracing sessions.
> Any user
> that wants to manage either a kernel trace using LTTng or an
> application trace
> with UST must interact with that registry for any possible actions.
>
We can start here... :-)
What do we need the session daemon for in the case where a single user
has downloaded
and compiled UST in his home directory? What is wrong with the socket
per app
and direct connection approach. Please clarify WHAT exactly we need a
session daemon for.
> The model address multiple tracing use cases based on the fact that we
> introduce a tracing Unix group (tracing group). Only users in that
> group or
> root can use the global registry. Other users will create a local
> registry
> (per-user registry) that will be completely independent from the
> global one.
>
> Two cases:
>
> 1) Users in the tracing group, it's tracing session can consume
> all tracing
> buffers from all applications and the kernel.
>
> 2) Users NOT in the tracing group, it's tracing session can only
> consume
> data from its own applications' buffers hence tracing his
> applications.
>
> A session stored by the registry consist of:
>
> * Session name (given by the user or automatically assigned)
> * List of traces (LTTng or UST)
> * Tracepoints/markers associated to a trace of that session
> * UUID
> * Associated user (UID)
>
> Then, consumers are used to extract data from tracing buffers. These
> consumers
> are daemon consuming either UST or/and LTTng buffers. For a single
> session,
> only one UST consumer and one LTTng consumer is necessary. The daemon CAN
> handle multiple tracing buffers for network streaming by example or
> for quick
> snapshot. These consumers are told by the inprocess library or the
> kernel to
> start getting out data on disk or network.
>
> For the next subsections, every components of this new proposal is
> explained
> from the global and per-user registry perspective.
>
> LTT-SESSIOND:
>
> The ltt-sessiond daemon acts as a session registry i.e. by keeping
> reference to
> all active session and, by active, it means a session in any state
> other than
> destroyed. Each entity we are keeping track of, here session, will have a
> universal unique identifier (UUID) assigned to it. The purpose of this
> UUID is
> to track a session in order to apply any kind of actions (Ex: Attach,
> Destroy).
> A human readable version SHOULD be consider in order to facilitate the
> session
> identification when listed by drtrace.
>
> The daemon creates two local Unix sockets (AF_UNIX). The first one is
> for what
> we call client communication i.e. interaction with drtrace (or any other
> compatible tools). That socket is set with the ltt-sessiond
> credentials with
> read-write mode for both user and group. The second one is a global
> socket for
> application registration for the UST case (see inprocess lib
> subsection below).
>
> This daemon is also responsible for tracing buffers creation. Two main
> reasons
> motivate this design:
>
> * The ltt-sessiond needs to keep track of all the shared memory
> segments in
> order to be able to give reference to any other possible consumer.
>
> * For the case of sharing tracing buffers between all userspace
> applications, having the registry allocating them will allow that
> but, if
> the inprocess library was allocating them, we will need to
> redesign the
> whole model.
>
Ok, here I once again need a motivation in the text.
App is user A
Session daemon user root with gid trace
consumer has gid trace.
Here is another approach that I'd like you to consider.
app creates buffer with mode 0700.
the session daemon is connected so the app sends a reference to the
session daemon containing the shmid.
The trace daemon changes the mode to 0750 and the gid to the tracing group
and then discards the reference. Hej presto, we don't need to keep a
reference
to the shared memory and the consumers can happily consume.
> For all tracing actions either to interact with a session or a
> specific trace,
> the drtrace client MUST go through ltt-sessiond. The daemon will take
> care of
> routing the command to the write inprocess library or the kernel.
>
> Global registry:
>
> A global registry SHOULD be started, idealy at boot, with credentials
> UID root
> and GID of the tracing group. Only user within the tracing group will
> be able
> to interact with that registry. All applications will try to register
> to that
> registry using the global socket (second one discuss above).
>
> Per-user registry:
>
> This type of registry address two use cases. The first one is when a
> session
> creation is requested from drtrace but no global ltt-sessiond exist.
> So, a
> ltt-sessiond will be spawned in order to manage the tracing of that
> user. The
> second use case is when a user is not in the tracing group thus he cannot
> communication with the global registry.
Ok, these are use-cases that assume a session daemon is a requirement
for tracing
at all, but not really a use-case for a session daemon.
>
> However, care MUST be put in order to manage the socket's daemon. They
> are not
> global anymore so they should be created in the home directory of the
> user
> requesting tracing.
>
> In both cases, for global and per-user registry, all applications MUST
> try to
> register to both ltt-sessiond. (see inprocess library subsection for
> details)
>
> The trace roles of ltt-sessiond:
>
> Trace interaction - Create, Destroy, Pause, Stop, Start, Set options
>
> Registry - keep track of trace's information:
> * shared memory location (only the keyid)
> * application PID (UST)
> * type (kernel or UST)
> * session name
> * UID
>
Can a non-priviled user trace the kernel?
Also, how aware are multiple session daemons of each other?
> Buffers creation - creates shared memory for the tracing buffers.
>
I really think buffer creation here is not a good idea.
I think you will end up putting state in way too many places and have to
replicate
information hither and thither in a way that will be very painful for you.
> UST-CONSUMERD:
Ok, we can start with a little note. ust-consumerd could just as well
have been
called ust-disk-consumer. It instantiates libustconsumer which does the
work.
I would want you to re-write this chapter taking into account multiple
instances
of libustconsumer.
>
> The purpose of this daemon is to consume the UST trace buffers for only a
> specific session. The session MAY have several traces for example two
> different
> applications. The client tool, drtrace has to create the ust-consumerd
> if NONE
> is available for that session. It is very important to understand that
> for a
> tracing session, there is only one ust-consumerd for all the traced
> applications.
So we are back at one libustconsumer instance per session? Why do we
want so many
processes hanging about?
>
> This daemon basically empty the tracing buffers when asked for and
> writes that
> data to disk for future analysis using LTTv or/and TMF (Tracing
> Monitoring
> Frameworks). The inprocess library is the one that tells the
> ust-consumerd
> daemon that the buffers are ready for consumption.
>
> Here is a flow of action to illustrate the ust-consumerd life span:
>
> 1)
> +-----------+ ops +--------------+
> | drtrace A |<---------->| ltt-sessiond |
> +-----------+ +--------------+
>
> drtrace ask for tracing an application using the PID and the session
> UUID. The
> shared memory reference is given to drtrace and the ust-consumerd
> communication
> socket if ust-consumerd already exist.
>
> 2a) If ust-consumerd EXIST
>
> +-----------+
> | drtrace A |
> +-----------+
> | mem ref.
> | +---------------+ read +------------+
> +-->| ust-consumerd |--------->| shared mem |
> +---------------+ +------------+
>
> In that case, drtrace only ask ust-consumerd to consume the buffers using
> the reference it previously got from ltt-sessiond.
>
> 2b) If ust-consumerd DOES NOT EXIST
>
> +-----------+ +--------------+
> | drtrace A | +---->| ltt-sessiond |
> +-----------+ | +--------------+
> | ID |
> | mem ref. | register
> | +---------------+
> +-->| ust-consumerd |
> +---------------+
>
> drtrace spawns the ust-consumerd for the session using the session
> UUID in
> order for the daemon to register as a consumer to ltt-sessiond for that
> session.
>
> Quick buffer snapshot:
>
> 1) Here, drtrace will request a buffer snapshot for an already running
> session.
>
> +-----------+ +--------------+
> | drtrace A |-------- ops ------->| ltt-sessiond |
> +-----------+ +--------------+
> | | command
> | +-----------------+ +-------+<--+
> | | ust-consumerd 1 |<----| app_1 |-+
> | +-----------------+ +-------+ | write
> | 1 | v
> | | +-------------+
> | +--- read ----->| shared mem. |
> | +-------------+
> | ^
> | +-----------------+ |
> +->| ust-consumerd 2 |----------+
> +-----------------+ snapshot
> | write
> |
> +---> disk/network
>
Hm, here you have the session daemon acting as a command router. I
really don't
recommend that.
I think I have to end here... But there is an old adage from somewhere
that goes
"show us the code". This form of programming on paper and design by
committee
is very painful, takes a lot of time and creates a lot of strife.
I recommend you put down the pen and start coding away instead. Put some
patches
up and we can review them and try them out. I think that you will
realize that
you want to minimize the amount of state that you replicate as well as
the total
amount of information you keep track of. Once you've done that we can start
discussing models again.
Even if we were to rubber stamp a final Daemon model proposal you can
very well
run into design issues along the way that force you to completely change
your
approach.
/Nils
> The first ust-consumerd (1) was already consuming buffers for the current
> session. So, drtrace ask for a live snapshot. A new ust-consumerd (2) is
> spawned, snapshot the buffers using the shared memory reference from
> ltt-sessiond, writes date to disk and die after all.
>
> On the security side, the ust-consumerd gets UID/GID from the drtrace
> credentials since it was spawned by drtrace and so the files
> containing the
> tracing data will also be set to UID/GID of the drtrace client. No
> setuid or
> setgid is used, we only use the credentials of the user.
>
> The roles of ust-consumerd:
>
> Register to ltt-sessiond - Using a session UUID and credentials
> (UID/GID)
>
> Consume buffers - Write data to a file descriptor (on disk,
> network, ...)
>
> Buffer consumption is triggered by the inprocess library which tells
> ust-consumerd when to consume.
>
> LTT-CONSUMERD:
>
> The purpose of this daemon is to consume the LTTng trace buffers for
> only a
> specific session.
>
> For that kernel consumer, ltt-sessiond will pass different anonymous file
> descriptors to the ltt-consumerd using a Unix socket. From these file
> desriptors, it will be able to get the data from a special function
> export by
> the LTTng kernel.
>
> ltt-consumerd will be managed by the exact same way as ust-consumerd.
> However,
> in order to trace the kernel, you are either root (UID=0) or in the
> tracing
> group.
>
> The roles of ltt-consumerd:
>
> Register to ltt-sessiond - Using a session UUID and credentials
> (UID/GID)
>
> Consume buffers - Write data to a file descriptor (on disk,
> network, ...)
>
> Kernel triggers ltt-consumerd for buffer consumption.
>
> UST INPROCESS LIBRARY:
>
> When the application starts, this library will check for the global
> named pipe
> of ltt-sessiond. If present, it MUST validate that root is the owner.
> This
> check is very important to prevent ltt-sessiond spoofing. If the pipe
> is root,
> we are certain that it's the privileged user that operates tracing.
> Then, using
> it's UID, the application will try to register to the per-user
> ltt-sessiond
> again verifying before the owner ship of the named pipe that should
> match the
> UID.
>
> Before registration, the inprocess library MUST validate with the
> ltt-sessiond
> the library version for compatibility reason. This is mechanism is
> useful for
> library compatibility but also to see if ltt-sessiond socket is valid
> (means
> that an actual ltt-sessiond is listening on the other side). Having no
> response
> for over 10 seconds, the application will cut communication on that
> socket and
> fallback to the application common named pipe (explain below).
>
> If the socket is valid, it will register as a traceable application
> using the
> apps credentials and will open a local Unix socket, passed to
> ltt-sessiond, in
> order to receive an eventual shared memory reference. It will then
> wait on it
> if any other command are given by the drtrace client. This socket
> becomes the
> only channel of communication between the registry and the application.
>
> If no ltt-sessiond is present at registration, the application tries
> to open
> the application common named pipe or create it if it does not exist
> and wait on
> it (using poll or epoll Linux API). Having any type of event on that
> pipe, the
> inprocess library will then try to register to the global and per-user
> ltt-sessiond. If it fails again, it goes back again to wait on that pipe.
>
> SHARED MEMORY
>
> For UST, this is the memory area where the tracing buffers will be
> held and
> given access in read-write mode for the inprocess library of the
> application.
>
> On the LTTng side (for ltt-consumerd), these buffers are in the kernel
> space
> and given access by opening a file in the debugfs file system. With an
> anonymous file desriptor, this consumer will be able to extract the data.
>
> This memory is ONLY used for the tracing data. No communication between
> components is done using that memory.
>
> A shared memory segment for tracing MUST be set with the tracing group
> GID for
> the UST buffers. This is the job of ltt-sessiond.
>
> PREREQUISITES:
>
> The global ltt-sessiond daemon MUST always be running as "root" or an
> equivalent user having the same privilege as root (UID = 0).
>
> The ltt-sessiond daemon SHOULD be up and running at all time in order
> to trace
> a tracable application.
>
> The new drtrace library API MUST be used to interact with the
> ltt-sessiond registry daemon for every trace action needed by the user.
>
> A tracing group MUST be created. Whoever is in that group is able to
> access the
> tracing data of any buffers and is able to trace any application or
> the kernel.
>
> WARNING: The tracing group name COULD interfere with other already
> existing
> groups. Care should be put at install time for that (from source and
> packages)
>
> The next section illustrates different use cases using that new model.
>
> Use Cases
> -----------------
>
> Each case considers these :
>
> * user A - UID: A; GID: A, tracing
> * user B - UID: B; GID: B, tracing
>
> Scenario 1 - Single user tracing app_1
> ------
>
> This first scenario shows how user A will start a trace for
> application app_1
> that is not running.
>
> 1) drtrace ask ltt-sessiond for a new session through a Unix socket. If
> allowed, ltt-sessiond returns a session UUID to the client.
> (Ex: ops --> new session)
>
> +-----------+ ops +--------------+
> | drtrace A |<---------->| ltt-sessiond |
> +-----------+ +--------------+
>
> 2) The app_1 is spawned by drtrace having the user A credentials.
> Then, app_1
> automatically register to ltt-sessiond has a "tracable apps" through
> the global
> named pipe of ltt-sessiond using the UID/GID and session UUID.
>
> The shared memory is created with the app_1 UID (rw-) and tracing
> group GID
> (r--) and a reference is given back to app_1
>
> +-----------+ +--------------+
> | drtrace A | | ltt-sessiond |
> +-----------+ +--------------+
> | ^ |
> | +-------+ | | +-------------+
> +-->| app_1 |<--------+ +-->| shared mem. |
> +-------+ +-------------+
>
> 3) app_1 connect to the shared memory and ust-consumerd is spawned
> with the
> session UUID and drtrace credentials (user A). It then register to
> ltt-sessiond
> for a valid session to consume using the previous session UUID and
> credentials.
>
> +-----------+ +--------------+
> | drtrace A | +-->| ltt-sessiond |----------+
> +-----------+ | +--------------+ |
> | | |
> | +---------------+ read | commands
> +-->| ust-consumerd |---------+ | and
> +---------------+ v | options
> ^ | +-------------+ |
> | v +------>| shared mem. | |
> +-------+ | +-------------+ |
> | app_1 |-------- |
> +-------+ write |
> ^ |
> +---------------------------------------
>
> Scenario 2 - Single user tracing already running app_1
> ------
>
> 1) drtrace ask ltt-sessiond for a new session through a Unix socket.
> If allowed
> (able to write on socket), ltt-sessiond returns a session UUID to the
> client.
>
> +-----------+ ops +--------------+
> | drtrace A |<---------->| ltt-sessiond |
> +-----------+ +--------------+
> ^
> +-------+ read |
> | app_1 |----------+
> +-------+
>
> NOTE: At this stage, since app_1 is already running, the registration
> of app_1
> to ltt-sessiond has already been done. However, the shared memory
> segment is
> not allocated yet until a trace session is initiated. Having no shared
> memory,
> the inprocess library of app_1 will wait on the local Unix socket
> connected to
> ltt-sessiond for the reference.
>
> +-----------+ +--------------+
> | drtrace A | | ltt-sessiond |
> +-----------+ +--------------+
> ^ |
> +-------+ | | +-------------+
> | app_1 |<--------+ +-->| shared mem. |
> +-------+ commands +-------------+
> | ^
> +---------- write ----------+
>
> 2) drtrace spawns a ust-consumerd for the session. We get the same
> figure as
> step 3 in the first scenario.
>
> There is a small difference though. The application MAY NOT be using
> the same
> credentials as user A (drtrace). However, the shared memory is always
> GID of
> the tracing group. So, in order for user A to trace app_1, is MUST be
> in the
> tracing group otherwise, if the application is not set with the user
> credentials, user A will not be able to trace app_1
>
> Scenario 3 - Multiple users tracing the same running application
> ------
>
> 1) Session are created for the two users. Using the same exact
> mechanism as
> before, the shared memory and consumers are created. Two users, two
> sessions,
> two consumers and two shared memories for the same application.
>
> +-----------+ +--------------+
> | drtrace A |-------- ops ------->| ltt-sessiond |
> +-----------+ ^ +--------------+
> | ^ commands
> +-----------+ | +-------+<--+
> | drtrace B |------+ +--->| app_1 |------- write -----+
> +-----------+ | +-------+ |
> | |
> +-----------------+ | +-------------+ |
> | ust-consumerd A |--O--- read ----->| shared mem. |<-+
> +-----------------+ | +-------------+ |
> | |
> +-----------------+ v +-------------+ |
> | ust-consumerd B |--+--- read ----->| shared mem. |<-+
> +-----------------+ +-------------+
>
> ust-consumerd A - UID: user A (rw-), GID: tracing (r--)
> ust-consumerd B - UID: user B (rw-), GID: tracing (r--)
>
> Scenario 4 - User not in the tracing group
> ------
>
> For this particular case, it's all goes back to the first scenario.
> The user
> MUST start the application using his credentials. The session will be
> created
> by the per-user ltt-sessiond but he will not be able to trace anything
> that the
> user does not owned.
^ permalink raw reply [flat|nested] 5+ messages in thread